Source · repo

Open Code Review (Alibaba)

Author
Alibaba
Published
17 November 2026
Added
17 November 2026
Original
https://github.com/alibaba/open-code-review

Summary

Open Code Review (OCR) is Alibaba's Apache-2.0 AI code review CLI. It mixes deterministic steps (choosing files, bundling them, matching rules, placing comments) with an LLM agent that writes line-level review comments on a diff or a whole directory. It ships as a CLI, as agent skills, and as plugins for Claude Code, Codex, Cursor, and others, and can run in CI. date_published is the date the README was read; the repo had about 53,000 stars and 870 commits, with no tagged release shown.

Key points

  • Install: npm install -g @alibaba-group/open-code-review (needs Git 2.41+), then ocr config provider and ocr config model to pick an LLM. As skills for any Agent Skills agent: npx skills add alibaba/open-code-review.
  • Use: ocr review reviews staged, unstaged, and untracked changes; ocr review --from main --to my-branch reviews a branch; ocr review --commit <sha> one commit; ocr scan --path <dir> whole files. --format json --output result.json for tooling; --resume <session-id> continues a review.
  • Two modes: native mode (open-code-review skill) uses OCR's own agent and model settings; delegation mode (open-code-review-delegate skill) lets your coding agent do the review with its own model and login, while OCR supplies file selection, rules, and report checks. Delegation needs no separate API key.
  • Rules: built-in rules for null dereferences, thread safety, XSS, and SQL injection; extensible with MCP tools. CI templates for GitHub Actions, GitLab CI, and Gerrit; OpenTelemetry tracing and a session viewer.
  • Claims: the README reports higher precision than Claude Code's review at about a ninth of the tokens, with lower recall. These are Alibaba's own benchmark figures.

Why it matters

A free, local review step that can run inside the coding agent you already use (delegation mode) or in CI, with deterministic rules to keep comments focused. Useful as a pre-PR check before a human or hosted review.